Repository path: workshop/experiments/E-20260903-latin-in-italian/design.md · rendered 2026-09-09

E-20260903-latin-in-italian — design, frozen before any English rendering of the work was opened

Frozen 2026-09-03 (S241), ARM-latin-in-italian step 1. The loci were enumerated from the Italian and classified (loci-frozen.md) before any of the four English volumes was read; the translation limb was drafted and frozen from the Italian alone. What had been read of the English side at freeze time is declared in §7.

1. Question

framework/v0.2 §7.45 is evidenced on Arabic inside Persian, one chapter, four hands. Does it reach Latin inside Italian — Dante's «Vita Nova», four published English hands, 1861–1902 — and does its three-way option set (mark / print plain / cut) survive a readership that can read the embedded language and therefore admits a fourth move, englishing the quotation?

2. Materials

Source. Dante Alighieri, «Vita Nova». Copy-text: Pesaro, Tipografia Nobili, 1829 (Internet Archive vitanovadidante00aliggoog), as transcribed at it.wikisource.org (Vita Nova/Vita nova, rev. 2026-05-06, own quality flag 25%), apparatus stripped programmatically, stored at materials/vn-it-body.txt. The 1829 scan's own OCR is stored at materials/raw/vitanovadidante00aliggoog.djvu.txt and is the collation witness. The italic in the copy-text is the 1829 editor's, not Dante's: the manuscripts carry no such distinction, which is exactly the fact that makes the English hands' italic a supplied signal rather than a carried one.

Hands. Four published English renderings of the whole work, italic recovered from each scan's ABBYY layer by abbyy_italics.py (the script E-20260831-arabic-in-persian used, unchanged):

Fifth, labelled, not published: the lead's own T-vita-nuova-R04-v1, XXIV–XXVI, which covers seven of the 21 loci. It is reported separately and is never pooled with the published hands.

3. Loci — enumerated from the Italian, two passes required to agree

enumerate.py runs both:

• (a) the 1829 editor's italic runs, carried out of the transcription as {{ }};
• (b) a mechanical Latin-inflection scan of the same text with the italic runs removed.

(a) alone would make the locus set depend on one editor's typography, which is the variable under study. Result: 21 loci by (a); residue of (b) is empty — no Latin is sitting in roman type in the copy-text. The frozen list, with each locus's classification, is loci-frozen.md.

Two codings, both read off the Italian only:

frame — does Dante name whose the words are? ATTRIB (an external author or source is named: «per quelle parole di Jeremia profeta», «appare per Virgilio») · SPEECH (a figure inside the narrative «disse queste parole», no external source named) · BARE (neither).

intro — does the Italian verbally announce that quoted words follow, in any form? Only two loci do not, and they are the two that stand inside Dante's sentence rather than beside it.

function — what is the Latin doing? SCRIPT (scripture or liturgy) · AUCT (a learned authority quoted as authority) · DRAMA (words spoken by a figure in Dante's own vision or narrative) · RUBRIC (a heading, or a formula fused grammatically into Dante's own sentence).

Frozen counts: frame ATTRIB 9 · SPEECH 8 · BARE 4; intro no 2. function SCRIPT 5 · AUCT 7 · DRAMA 7 · RUBRIC 2.

4. Coding scheme — one cell per locus × hand

Two further binary marks per cell, both read off the English page only:

• verbal — the hand adds a verbal signal the Italian does not have ("in Latin", "the Latin runs", "saith the prophet") at or beside the locus.
• gloss — the hand supplies an English rendering in addition to a retained Latin run (a doubling the Italian does not have).

Every coded cell is quoted verbatim in coding.md, with the page as the ABBYY layer gives it, so a second reader can check the call without re-fetching the scans.

Device availability control (§7.45.4). A hand that marks at 0 loci is only evidence of a policy if the volume uses italic at all. For any such hand the volume is searched for italic runs outside the loci, and the finding is stated on the result page.

5. Registered predictions

Registered before any of the four English volumes was read. Each is a transfer of a printed framework sentence, except P3, which is a genuine open bet the Persian material could not carry.

• P1 — the signal is supplied here too (§7.45.1). Among retained Latin runs, pooled over the four hands, the italic share is ≥ 0.50; and at least one hand marks at ≥ 0.80 of its own retained runs.
• P2 — the signal is about the language, not the status (§7.45.5). For each hand that marks at all, |rate(DRAMA) − rate(SCRIPT)| ≤ 0.15, the rate being italic-among-retained.
• P3 — the fourth option is the minority move. Pooled over hands and loci, ENGLISHED + ENG+NOTE ≤ 0.25 of non-NA cells. This is the prediction with no framework behind it: the handbook has never met a target readership that could read the embedded language.
• P4 — the unattributable quotation (§7.45.6). No locus is DELETED by all four hands; and if one is, it is L10 (Nomina sunt consequentia rerum), the only locus with neither a named author nor a named speaker.

P1, P2, P4 are the framework's; a failure of any of them is a limit on §7.45's printed scope, not a defect of the Persian result. P3 is an open bet and is reported whichever way it falls.

6. Failure criteria — what would make the run uninterpretable

1. Alignment. Fewer than three hands align at ≥ 18 of 21 loci. (A hand that abridges is coded NA at the missing loci and reported, not discarded, but three aligned hands is the floor for any pooled rate.)
2. Italic recovery. If the ABBYY layer for a volume yields no italic anywhere, that volume's marking cells are void and the hand is reported for its keep/english/delete behaviour only.
3. Locus disagreement. If pass (b) had returned any Latin outside the editor's italic, the locus set would have been rebuilt from (b) before coding. It returned none.

7. What had been read of the English side at freeze time

Declared, because the freeze is only worth what it declares. Before the loci were frozen and before the translation was drafted, the lead had read: the Internet Archive metadata of all four volumes; and the introduction (pp. i–xii) of vitanuova00aliggoog, an 1892 Chiswick Press edition of the Italian text with English editorial matter, which was examined to see whether it was a translation and set aside when it proved not to be. It contains no English rendering of the «Vita Nova» prose. No page of ROS, MAR, NOR or MEY was opened before workshop/translations/vita-nuova/R04-v1/translation.md and its log were committed.

8. Cost

$0. No API call. The translating is the lead's own (charter §3, A4); the four volumes and the Italian source are public domain and were fetched directly.

9. What this design cannot do

• One reader codes every cell. The coding is internal-judgment-only, and its defence is that every cell is quoted. §7.45's Persian coding has the same limit and the same defence.
• It is observational. Nothing is assigned; four hands chose their own policies, and three of the four wrote prefaces that may state them. Where a preface states a policy it is quoted, and a policy stated is evidence about the hand, never about translation in general.
• MEY 1902 and the two 1860s hands are not independent of each other in the way the arithmetic pretends. Later translators of a famous work read earlier ones. Any pooled rate is a rate over four books, not over four independent draws.
• The lead's seven cells are not a fifth hand in any statistical sense — one translator, one span, and the same agent that wrote this design.